Shifts from your sales
Paste an hourly sales export (a date-time column and a sales column). Each weekday-hour is averaged, heads needed is sales divided by your sales per labor hour (at least your minimum), and each day's headcount is cut into shifts. Copy the result into the roster builder. Square and Toast item exports work as they come (checked on real ones): net sales are used, and voided or refunded lines are left out. Square exports one date range at a time, so you can load several files at once: each file's columns are matched by name and a row two files share is counted once. Nothing is uploaded.
Your export has several shops in it (a Location, Store or Site column, as Square and Toast exports have). Plan them as one, or pick one to plan it alone.
Blend any of the three, e.g. 2/ticket + 0.05/sale: the minutes a ticket takes at the till, a unit of money (bigger orders, more to pack) and an item (each thing picked). The rate is then minutes of work per labor hour: 60 is someone busy every minute, 45 leaves a quarter of each hour for cleaning and restocking. Work that does not grow with customers can be added as fixed minutes: 10/hour in every hour you trade, 30/open to set up in the first hour of the day and 45/close to clean down in the last. If that work happens with the doors shut, write 20/before-open or 45/after-close: it goes in the hour before you open or after you close, so the opener comes in early or the closer stays on, and minimum staff does not apply to that hour. With the doors shut nobody waits on a customer, so that work is planned at 60 minutes a person and runs on into the next hour: 90/after-close is one person staying two hours, not two people for one.
The fewest people on in an hour of before-open or after-close work. Put 2 if nobody may open or lock up alone; the two then share the work. It can differ by weekday: 1, Fri/Sat=2 is one closer most nights and two on late Fridays and Saturdays.
auto tests equal weights, half-lives of 8, 4 and 2 weeks and the last 4 weeks only on your own file (each past week planned from the weeks before it) and keeps whichever left the fewest hours short or spare. It needs 8 or more full weeks of dated sales; on a long file it takes a few seconds.
On (the default): minimum staff is needed only in hours your shop normally trades (sales on 1 in 5 of that weekday's dates or more) or sold in that day, so a plan that staffs a lunch closure or old closing hours is charged for it, and auto may pick the last 4 weeks only, which follows a change in opening hours fastest. Off: minimum staff counts as needed in every hour a plan staffs, and auto chooses among the weightings only.
Planned from the average, an hour is short about half the time. With a number here each hour is staffed for that share of its past days instead: 1 staffs for the median day, 3 for 3 days in 4 (75%), 0.5 for 1 in 3.
An extra person wanted for only an hour or two is checked against your past days before it becomes a whole shift. It goes in the cheapest shift that holds it, or is left out (and listed) when that shift's spare hours cost more than the short hours it saves, at this number or at 1 when blank. Minimum staff is always scheduled.
Shifts to schedule
Test it on your past weeks
Each full week after the first four is planned from the weeks before it only, then compared hour by hour with what that week's sales needed. Tried for the plain average, three half-lives, the last four weeks and last week alone; with a short-cost number, each of those staffed for that share of past days too, scored as that many times the short hours plus the spare ones. Each plan is cut into shifts first, as above, so the hours added to reach the shortest shift count; with a short-cost number, a run shorter than the shortest shift is staffed only where that is cheaper than leaving it short. Needs dated rows and at least five full weeks.